Runway Model Selection – Choosing the Right Generation Model
Lesson 3: Building a Model Testing Workflow
Lesson Objectives
By the end of this lesson, students should be able to:
- Design a systematic model test for a specific creative task
- Document model test results effectively
- Build a test library as a production reference
- Update the test library when new models become available
Lesson Content
Designing a model test.
An effective model test: uses representative content that matches the actual creative task (not generic samples), tests all relevant models with the same prompt to enable direct comparison, evaluates each dimension that matters for the task (quality, motion, cost), and generates at least two or three examples per model to account for generation variability.
Documenting test results.
Record: which models were tested, the specific prompt used, the evaluation dimensions and ratings, and which model was selected and why. This documentation is useful when facing similar creative tasks in future projects – reducing the need to re-run tests from scratch each time.
Building a test library.
Over time, model test results across different content types (outdoor environments, human subjects, abstract visuals, camera motion types) build a reference library – a practical asset for quickly identifying which model suits an upcoming project type without a full re-test, based on documented prior performance.
Updating for new model releases.
When Runway releases new models, run new comparative tests against your established test prompt set to update the library – adding the new model's performance data and adjusting production recommendations accordingly. The test library is a living reference, not a static one.
Enroll to continue this lesson.
The preview above shows the lesson objectives and opening lesson content. Enroll to view the full lesson, complete the practice work, and take the lesson quiz.